Qwen3.8-27B Long-Context Decoding Speeds Up Over 3x

Community-shared optimizations using DFlash2 plus XQA boost Qwen3.8-27B decoding from 26 tok/s to 92.9 tok/s at 192K context on an RTX PRO 6000 Max Q, a roughly 3.5x speedup.

2026-08-22 ~ 2026-08-23 · 2 related posts

• Qwen3.8-27B: 3x Faster Long-Context Decoding with DFlash2 + XQA — stargate425 · 2026-08-22
• Qwen 3.8 27B Decoding Speed Tripled to 92.9 tok/s via Optimization — TheMoonMidas · 2026-08-23